Scalable Parallel Debugging with g-Eclipse

Simulation software on today’s HPC systems needs to be scalable to a large number of processes to make efficient use of such machines. Already at this level and especially with the expected increasing scalability of upcoming machines, the development and debugging of parallel programs becomes an increasingly difficult task. Consequently, sophisticated tools providing mechanisms for handling large-scale parallel and distributed programs are needed. In this paper we show several ways to improve the handling of large event traces using the Trace Viewer plug-in of the g-Eclipse tool and we propose the use of a pattern matching technique to simplify the debugging of large message passing parallel programs. With the pattern matching approach, we enable an additional layer of abstraction, which supports the user in understanding the program’s behaviour.